Prepared by Johns Hopkins ID Dynamics Working Group
Updated 2020-07-08
FOR PLANNING PURPOSES ONLY: NOT A FORECAST
We compared 3 intervention scenarios for model simulations from January 31, 2020 through December 31, 2020:
[CHANGE BELOW DETAILS ABOUT INTERVENTIONS]
Lockdown followed by Worst-Case Uncontrolled Spread: This scenario has statewide school closures from March 13-19 followed by a statewide stay-at-home policy from March 19 through May 14. All interventions are then removed (back to worst-case uncontrolled transmission) starting May 15.
Lockdown followed by Test and Isolate: This scenario has statewide school closures from March 13-19 followed by a statewide stay-at-home policy from March 19 through May 14. From May 15 through December 31, there is a targeted test and isolate strategy similar to that implemented in South Korea.
Lockdown followed by Moderate Social Distancing: This scenario has statewide school closures from March 13-19 followed by a statewide stay-at-home policy from March 19 through May 14. From May 15 through December 31, there is moderately restrictive social distancing similar to that in US cities during the 1918 influenza pandemic.
[MAYBE WE SHOULD THEN ADD SOME SUMMARY STATEMENTS ABOUT OUR FINDINGS?]
[ALTER BELOW DESCRIPTIVE STRING FOR SCENARIO CAPTIONS]
Fig. 1: Daily hospital occupancy for 15 simulation realizations for three scenarios with 8-week lockdown followed by 1) worst-case uncontrolled spread, 2) targeted testing and isolation, and 3) moderately restrictive social distancingat 1 % IFR assumptions.
The model estimates the potential impact of non-pharmaceutical intervention scenarios that governments are using in response to the COVID-19 pandemic. By accounting for factors that help explain how the coronavirus that causes COVID-19 spreads through communities, such as volume of travel in and out of the county and movement within the county, the model provides reasonable estimates about how health outcomes (e.g., epidemic peak timing, hospital bed needs) may be impacted by each of the non-pharmaceutical interventions. These health impact estimates reflect information about local resources and capacity. The models are not predictions, but rather our best approximation of how the COVID-19 pandemic could play out under different government actions. As such, they should be used as decision making tools only; these estimates are not meant to be shared outside of decision-making contexts.
What is the difference?
The difference is important. This report presents planning models tailored to a specific location, with the aim of providing a best estimate of what may happen under different interventions. In contrast, forecasting models attempt to predict what will happen.
Why does that difference matter?
Planning models are a tool for decision makers who are responsible for making decisions about the jurisdictions they serve. They are less useful for members of the general public who are not making decisions for their cities. The different scenarios used in planning models assume that the government actions will result in large-scale behavior change that is best accomplished through government action, and can inform resource allocation decisions made by governments and healthcare organizations.
Tab.1: Summary acrossthree scenarios with 8-week lockdown followed by 1) worst-case uncontrolled spread, 2) targeted testing and isolation, and 3) moderately restrictive social distancing. Estimates are presented across4time periods for 1% infection fatality ratio (IFR) assumptions.
Jan 30-Jun 01 | Jan 30-Jun 01 | Jun 01-Jul 01 | Jun 01-Jul 01 | Jul 01-Aug 01 | Jul 01-Aug 01 | Aug 01-Jan 30 | Aug 01-Jan 30 | |
mean | 95% PI | mean | 95% PI | mean | 95% PI | mean | 95% PI | |
Infections in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 159,400 | 18,640-416,430 | 486,150 | 213,210-635,770 | 156,410 | 14,190-427,310 | 23,600 | 190-100,430 |
Lockdown followed by Test and Isolate | 109,490 | 14,310-232,030 | 467,980 | 189,830-639,820 | 205,500 | 38,970-459,930 | 26,540 | 620- 99,700 |
Lockdown followed by Worst Case Uncontrolled Spread | 369,500 | 55,270-661,520 | 340,210 | 190,470-519,890 | 96,180 | 2,930-333,080 | 10,770 | 20- 45,110 |
Deaths in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 380 | 30-1,300 | 3,710 | 710-6,060 | 3,430 | 1,130-5,890 | 760 | 10-3,030 |
Lockdown followed by Test and Isolate | 190 | 40- 470 | 3,120 | 520-5,470 | 3,840 | 2,450-4,940 | 910 | 50-3,000 |
Lockdown followed by Worst Case Uncontrolled Spread | 940 | 120-1,830 | 4,830 | 1,590-6,530 | 2,010 | 290-4,400 | 400 | 0-1,610 |
Hospital Admissions in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 10,050 | 1,150-29,530 | 48,650 | 17,360-68,600 | 20,310 | 2,800-44,850 | 3,540 | 50-14,800 |
Lockdown followed by Test and Isolate | 6,400 | 920-14,250 | 44,880 | 13,280-67,760 | 25,740 | 7,300-47,360 | 4,080 | 120-14,900 |
Lockdown followed by Worst Case Uncontrolled Spread | 27,600 | 3,460-52,010 | 40,070 | 25,560-55,610 | 12,360 | 630-38,760 | 1,640 | 10- 6,800 |
Peak Hospital Occupancy in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 6,310 | 670-16,530 | 26,190 | 10,660-34,270 | 19,200 | 7,640-30,820 | 3,870 | 120-14,520 |
Lockdown followed by Test and Isolate | 4,470 | 560- 9,610 | 22,470 | 8,560-32,200 | 20,250 | 16,470-22,780 | 4,970 | 350-15,380 |
Lockdown followed by Worst Case Uncontrolled Spread | 18,230 | 2,330-34,220 | 26,160 | 15,050-35,040 | 10,530 | 2,070-18,210 | 2,240 | 20- 8,780 |
ICU Admissions in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 1,190 | 100- 4,130 | 11,840 | 2,280-19,110 | 10,850 | 3,700-18,550 | 2,430 | 60- 9,510 |
Lockdown followed by Test and Isolate | 620 | 120- 1,470 | 10,010 | 1,720-17,650 | 12,290 | 7,960-15,960 | 2,990 | 170- 9,870 |
Lockdown followed by Worst Case Uncontrolled Spread | 2,970 | 350- 5,700 | 15,430 | 5,070-20,690 | 6,430 | 900-14,060 | 1,280 | 10- 5,130 |
Peak ICU Occupancy in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 960 | 80- 3,340 | 8,980 | 1,940-14,650 | 12,110 | 8,210-14,900 | 3,480 | 420- 7,990 |
Lockdown followed by Test and Isolate | 490 | 80- 1,150 | 7,770 | 1,450-13,480 | 11,250 | 8,630-14,300 | 4,650 | 1,100- 8,760 |
Lockdown followed by Worst Case Uncontrolled Spread | 2,690 | 290- 5,300 | 10,960 | 3,880-15,440 | 8,530 | 6,160-10,990 | 2,200 | 100- 7,250 |
Incident Ventilations in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 50 | 0- 4,130 | 780 | 100-19,110 | 2,380 | 1,130-18,550 | 770 | 50- 9,510 |
Lockdown followed by Test and Isolate | 20 | 0- 1,470 | 550 | 70-17,650 | 2,310 | 1,070-15,960 | 990 | 150- 9,870 |
Lockdown followed by Worst Case Uncontrolled Spread | 60 | 10- 5,700 | 1,810 | 290-20,690 | 1,610 | 800-14,060 | 460 | 10- 5,130 |
Peak Ventilators in Use in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 40 | 0- 160 | 630 | 80-1,410 | 1,800 | 890-2,250 | 1,110 | 320-1,970 |
Lockdown followed by Test and Isolate | 20 | 0- 50 | 470 | 60- 980 | 1,590 | 820-2,170 | 1,200 | 790-1,540 |
Lockdown followed by Worst Case Uncontrolled Spread | 50 | 10- 90 | 1,370 | 250-2,400 | 1,780 | 1,180-2,360 | 620 | 80-1,310 |
Fig. 1 : Distribution of cumulative hospital admissions for three scenarios with 8-week lockdown followed by 1) worst-case uncontrolled spread, 2) targeted testing and isolation, and 3) moderately restrictive social distancing at 1% IFR assumptions. Each bar represents a bin of 2,000 hospitalizations.
Tab. 2 State-level summary for Lockdown followed by Worst Case Uncontrolled Spread scenario reported for the period from 2020-1-31 through the dates specified by the column headers.
| IFR | Jun 01 |
| Jul 01 |
| Aug 01 |
|
INFECTIONS |
| 387,100 | (60,400-683,400) | 715,500 | (407,500-852,400) | 806,800 | (728,300-854,900) |
HOSPITALIZATIONS | 0.25% IFR | 7,400 | ( 900-13,800) | 17,100 | ( 8,100-21,100) | 20,000 | (17,600-21,300) |
| 0.5% IFR | 14,800 | (1,900-27,900) | 34,100 | (16,300-42,700) | 40,000 | (35,300-43,000) |
| 1% IFR | 29,500 | (3,800-55,100) | 68,400 | (32,900-84,800) | 80,200 | (70,700-85,300) |
daily peak admissions | 0.25% IFR | 500 | (100- 900) | 700 | ( 400- 900) | 700 | ( 400- 900) |
| 0.5% IFR | 1,000 | (200-1,700) | 1,300 | ( 800-1,700) | 1,300 | ( 900-1,700) |
| 1% IFR | 2,000 | (400-3,400) | 2,600 | (1,600-3,400) | 2,600 | (1,800-3,400) |
daily peak capacity | 0.25% IFR | 4,700 | ( 600- 8,700) | 6,600 | ( 3,800- 8,800) | 6,800 | ( 4,700- 8,800) |
| 0.5% IFR | 9,400 | (1,300-17,600) | 13,200 | ( 7,700-17,700) | 13,600 | ( 9,500-17,700) |
| 1% IFR | 18,900 | (2,600-34,800) | 26,300 | (15,500-35,000) | 27,000 | (18,800-35,000) |
ICU | 0.25% IFR | 900 | (100-1,600) | 4,700 | (1,400- 6,600) | 6,200 | ( 4,900- 6,800) |
| 0.5% IFR | 1,700 | (200-3,300) | 9,400 | (2,900-13,400) | 12,500 | ( 9,900-13,800) |
| 1% IFR | 3,400 | (400-6,500) | 18,700 | (5,800-26,500) | 24,900 | (19,800-27,300) |
daily peak admissions | 0.25% IFR | 100 | (10-200) | 200 | (100- 300) | 200 | (200- 300) |
| 0.5% IFR | 200 | (20-400) | 400 | (200- 600) | 400 | (300- 600) |
| 1% IFR | 400 | (40-800) | 800 | (400-1,100) | 900 | (600-1,100) |
daily peak capacity | 0.25% IFR | 800 | (100-1,500) | 2,700 | (1,000- 3,800) | 3,000 | (2,200- 3,800) |
| 0.5% IFR | 1,500 | (200-3,000) | 5,600 | (2,000- 7,800) | 6,100 | (4,300- 7,800) |
| 1% IFR | 3,000 | (300-6,000) | 11,000 | (4,100-15,400) | 12,000 | (8,600-15,400) |
DEATHS | 0.25% IFR | 300 | ( 30- 500) | 1,500 | ( 500-2,100) | 2,000 | (1,600-2,100) |
| 0.5% IFR | 500 | ( 60-1,000) | 2,900 | ( 900-4,100) | 3,800 | (3,100-4,200) |
| 1% IFR | 1,100 | (130-2,100) | 5,900 | (1,800-8,300) | 7,800 | (6,200-8,500) |
Fig. 2 County-level map of infections per 10,000 for Lockdown followed by Worst Case Uncontrolled Spread strategy.
Tab. 3 State-level summary for Lockdown followed by Test and Isolate scenario reported for the period from 2020-1-31 through the dates specified by the column headers.
| IFR | Jun 01 |
| Jul 01 |
| Aug 01 |
|
INFECTIONS |
| 121,200 | (15,800-254,000) | 589,500 | (218,500-810,200) | 785,100 | (671,900-847,900) |
HOSPITALIZATIONS | 0.25% IFR | 1,800 | ( 300- 4,000) | 13,000 | ( 3,700-19,400) | 19,100 | (15,400-21,000) |
| 0.5% IFR | 3,600 | ( 500- 8,000) | 26,200 | ( 7,700-39,100) | 38,500 | (31,300-42,500) |
| 1% IFR | 7,100 | (1,000-15,800) | 52,700 | (15,300-78,500) | 77,300 | (62,700-85,000) |
daily peak admissions | 0.25% IFR | 200 | (20- 400) | 600 | ( 300- 800) | 600 | ( 500- 800) |
| 0.5% IFR | 400 | (50- 800) | 1,100 | ( 600-1,600) | 1,200 | (1,000-1,600) |
| 1% IFR | 800 | (90-1,600) | 2,300 | (1,100-3,100) | 2,400 | (1,900-3,100) |
daily peak capacity | 0.25% IFR | 1,300 | (200- 2,700) | 5,600 | (2,300- 7,900) | 6,200 | ( 5,000- 7,900) |
| 0.5% IFR | 2,500 | (300- 5,400) | 11,200 | (4,600-16,000) | 12,600 | (10,300-16,000) |
| 1% IFR | 5,000 | (600-10,700) | 22,700 | (9,200-32,200) | 25,400 | (20,400-32,200) |
ICU | 0.25% IFR | 200 | ( 30- 400) | 2,800 | ( 500- 4,900) | 5,700 | ( 3,600- 6,700) |
| 0.5% IFR | 400 | ( 60- 800) | 5,600 | (1,000- 9,900) | 11,600 | ( 7,600-13,600) |
| 1% IFR | 700 | (120-1,600) | 11,200 | (2,000-19,900) | 23,100 | (15,000-27,100) |
daily peak admissions | 0.25% IFR | 20 | (10- 40) | 200 | ( 30- 300) | 200 | (200- 300) |
| 0.5% IFR | 30 | (10- 70) | 300 | ( 90- 500) | 400 | (300- 500) |
| 1% IFR | 60 | (10-130) | 600 | (170-1,000) | 800 | (600-1,000) |
daily peak capacity | 0.25% IFR | 100 | (20- 300) | 2,000 | ( 300- 3,400) | 2,800 | (2,200- 3,500) |
| 0.5% IFR | 300 | (50- 700) | 4,000 | ( 800- 6,900) | 5,600 | (4,500- 7,100) |
| 1% IFR | 500 | (90-1,200) | 8,000 | (1,600-13,800) | 11,300 | (8,800-14,300) |
DEATHS | 0.25% IFR | 100 | (10-100) | 900 | (100-1,500) | 1,800 | (1,200-2,100) |
| 0.5% IFR | 100 | (20-200) | 1,800 | (300-3,100) | 3,600 | (2,400-4,300) |
| 1% IFR | 200 | (40-500) | 3,500 | (600-6,200) | 7,200 | (4,800-8,400) |
Fig. 3 County-level map of infections per 10,000 for Lockdown followed by Test and Isolate strategy.
Tab. 5 : Summary across three scenarios with 8-week lockdown followed by 1) worst-case uncontrolled spread, 2) targeted testing and isolation, and 3) moderately restrictive social distancing for the county of New Castle . Estimates are presented across 4 time periods for 1% IFR assumptions.
Jan 30-Jun 01 | Jan 30-Jun 01 | Jun 01-Jul 01 | Jun 01-Jul 01 | Jul 01-Aug 01 | Jul 01-Aug 01 | Aug 01-Jan 30 | Aug 01-Jan 30 | |
mean | 95% PI | mean | 95% PI | mean | 95% PI | mean | 95% PI | |
Infections in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 66,900 | 4,100-168,400 | 303,600 | 87,900-395,100 | 105,700 | 11,700-289,100 | 19,500 | 200- 83,900 |
Lockdown followed by Test and Isolate | 31,900 | 7,300- 75,600 | 293,200 | 117,600-421,500 | 145,100 | 31,800-277,600 | 15,800 | 500- 55,400 |
Lockdown followed by Worst Case Uncontrolled Spread | 184,300 | 17,500-362,100 | 225,000 | 143,500-341,400 | 71,600 | 2,300-244,000 | 8,800 | 0- 36,900 |
Deaths in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 0 | 0- 100 | 1,100 | 100-2,000 | 1,100 | 500-1,800 | 300 | 0-1,200 |
Lockdown followed by Test and Isolate | 0 | 0- 0 | 700 | 200-1,500 | 1,400 | 1,000-1,800 | 300 | 0- 900 |
Lockdown followed by Worst Case Uncontrolled Spread | 100 | 0- 300 | 1,400 | 300-2,100 | 700 | 100-1,300 | 200 | 0- 700 |
Hospital Admissions in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 1,700 | 100- 4,500 | 15,000 | 3,200-21,000 | 6,600 | 1,100-14,000 | 1,400 | 0- 6,100 |
Lockdown followed by Test and Isolate | 800 | 200- 1,900 | 13,000 | 4,100-20,900 | 9,200 | 3,000-14,500 | 1,200 | 0- 4,200 |
Lockdown followed by Worst Case Uncontrolled Spread | 6,300 | 500-13,200 | 13,000 | 5,700-17,500 | 4,500 | 200-13,700 | 700 | 0- 2,900 |
Peak Hospital Occupancy in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 1,400 | 100- 3,500 | 8,300 | 2,200-10,600 | 6,600 | 3,200- 9,700 | 1,500 | 100- 5,600 |
Lockdown followed by Test and Isolate | 600 | 100- 1,500 | 7,300 | 2,700-10,700 | 7,200 | 6,200- 8,600 | 1,600 | 100- 4,400 |
Lockdown followed by Worst Case Uncontrolled Spread | 4,800 | 400-10,100 | 8,200 | 3,700-11,500 | 3,800 | 800- 6,200 | 900 | 0- 3,500 |
ICU Admissions in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 100 | 0- 300 | 3,400 | 400-6,200 | 3,500 | 1,500-5,800 | 1,000 | 0-3,800 |
Lockdown followed by Test and Isolate | 100 | 0- 200 | 2,400 | 500-4,800 | 4,400 | 3,100-5,900 | 900 | 100-2,800 |
Lockdown followed by Worst Case Uncontrolled Spread | 500 | 0-1,000 | 4,600 | 1,000-7,000 | 2,200 | 400-4,300 | 500 | 0-2,100 |
Peak ICU Occupancy in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 100 | 0- 300 | 2,700 | 300-4,500 | 3,900 | 2,500-4,600 | 1,200 | 100-2,800 |
Lockdown followed by Test and Isolate | 100 | 0- 100 | 2,000 | 400-4,000 | 3,800 | 2,700-4,700 | 1,700 | 500-2,700 |
Lockdown followed by Worst Case Uncontrolled Spread | 500 | 0-1,000 | 3,400 | 800-5,000 | 3,000 | 2,400-3,500 | 800 | 0-2,600 |
Incident Ventilations in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 0 | 0- 300 | 200 | 0-6,200 | 700 | 200-5,800 | 300 | 0-3,800 |
Lockdown followed by Test and Isolate | 0 | 0- 200 | 100 | 0-4,800 | 700 | 300-5,900 | 300 | 0-2,800 |
Lockdown followed by Worst Case Uncontrolled Spread | 0 | 0-1,000 | 500 | 0-7,000 | 500 | 300-4,300 | 200 | 0-2,100 |
Peak Ventilators in Use in Period | ||||||||
Lockdown followed by Moderate Social Distancing | 0 | 0- 0 | 200 | 0-400 | 600 | 200-700 | 400 | 100-600 |
Lockdown followed by Test and Isolate | 0 | 0- 0 | 100 | 0-200 | 500 | 300-700 | 400 | 300-600 |
Lockdown followed by Worst Case Uncontrolled Spread | 0 | 0- 0 | 400 | 0-800 | 600 | 300-800 | 200 | 0-400 |
The model results here are the results of a modeling pipeline that moves from a seeding phase, through epidemic simulation, a outcome and hospitalization generation engine, and report generation. Core code for all aspects of the pipeline is all open source. The core pipeline phases are:
Seeding: We run a model to determine the likelihood of importing cases into the region of interest. We maintain an R package.
Epidemic Simulation: We run a location stratified SEIR model, using a python module in COVIDScenarioPipeline.
Hospitalization and Outcome Generator: We estimate secondary effects of infection using infection numbers using the hospitalization R package in COVIDScenarioPipeline.
Report Generation: We provide functions to access results and help produce reports in the report_generation R package in COVIDScenarioPipeline.
County-level confirmed SARS-COV-2 infections: JHU CSSE COVID-19 Data Portal
US Census Bureau 2010 county commuting flows and 2016 population data
Our model and report make the following key assumptions:
Mean incubation period: 5.2 days
Infectious period: ranges from 2.6-6 days
R0: 2-3
We examine 3 infection fatality rates (IFR) throughout our report: 0.25, 0.5, 1%.
We assume that 10% of all hospitalized patients will die and adjust overall hospitalization rates to match the given IFR for a scenario.
Hospitalizations admitted to the ICU: 32%
ICU admissions that are ventilated: 15%
[CHOOSE ONE OF THESE CHUNKS DEPENDING ON SEEDING. POISSON SEEDING TEXT SAYS 10X] <!–
We used a global air importation model in conjunction with global cumulative confirmed COVID case data from JHU CSSE COVID-19 Data Portal and assumptions in recent travel reductions to estimate the number of COVID case importations into major airports located in and surrounding areas (Delaware) from January through March. These estimated global importations were used to seed our epidemic model.
If the number of infections in any single county was below the number of confirmed cases reported by the JHU CSSE COVID-19 Data Portal on a given day between February 1 and March 18, the simulation was stopped and removed from the model results. We ran 1000 model simulations from January 31, 2020 through December 31, 2020 for each scenario. Some outcomes may be reported beyond the simulation period in order to examine the full disease course for infections occurring at the end of the simulation period.
We seeded the epidemic with roughly 10x the number of confirmed cases reported by the JHU CSSE COVID-19 Data Portal for the five days prior to the first confirmed cases into each county in Louisiana. We ran 1000 model simulations from January 31, 2020 through December 31, 2020 for each scenario. Some outcomes may be reported beyond the simulation period in order to examine the full disease course for infections occurring at the end of the simulation period.
The epidemiological model is an stochastic spatial ‘SEIR model’ where the population of each node is divided in categories depending on their situation with regard to the diseases. The compartments are:
In practice, this translate into \(k = 3\) compartments of infected, \(I_1, I_2, I_3\) in order to have an infectious period shaped as Erlang distribution. Then the rates of the different compartments are given in the table below:
| Transition | rate parameter | Unit |
| \(S\longrightarrow E\) | \(\beta = R_0 \cdot \gamma\) d\(^{-1}\) | |
| \(E\longrightarrow I_1\) | \(\sigma = \frac{1}{5.2}\) | d\(^{-1}\) |
| \(I_1\longrightarrow I_2\) | \(\gamma_1 = \gamma \cdot k\) | d\(^{-1}\) |
| \(I_2\longrightarrow I_3\) | \(\gamma_1 = \gamma \cdot k\) | d\(^{-1}\) |
| \(I_3\longrightarrow R\) | \(\gamma_1 = \gamma \cdot k\) | d\(^{-1}\) |
For such epidemiological model, two meta-parameters are important in reproducing the dynamics and final size of an epidemic: the serial interval \(SI\), which is the interval between two subsequent infections, and the basic reproductive number \(R_0\) representing the number of newly infected per infected, in a fully susceptible population.
The serial interval (SI) of COVID-19 is currently estimated to be in range \(6.5-8.2\). We draw uniformly from this range, and we solve \(SI = \frac{1}{2}(\frac{1}{\gamma})+\frac{1}{\sigma}\) for the inverse of the total infectious period, \(\gamma\).
The basic reproductive number, \(R_0\), is drawn for each simulation and we obtain parameter beta from \(\beta= R_0 \cdot \gamma\).
The model is fixed time step, and the transitions (without mobility here) at each time step \(\Delta t\) are:
\[\begin{eqnarray} p_{expose} &=& 1 - \exp(-\Delta t \cdot FOI) \\ p_{infect} &=& 1 - \exp(-\Delta t \cdot \sigma)\\ p_{recover} &=& 1 - \exp(-\Delta t \cdot \gamma) \end{eqnarray}\] At each time step, we draw in a binomial distribution, e.g \[\begin{equation} N_{I_1\longrightarrow I_2}(t) = \text{Binom}(I_1, 1 - \exp(-\Delta t \cdot \gamma_1)) \end{equation}\] The force of infection without mobility is defined as: \[\begin{equation} FOI = \beta \frac{(I_1 + I_2 + I_3)^\alpha}{H} \end{equation}\]where \(\alpha\) is a coefficient that dampens disease transmission when its value is below 1, thus more accurately representing non-homogeneous mixing in the population. Although we were unable to find \(\alpha\) estimates for SARS-CoV-2, we expect disease transmission to be similar to that of other respiratory diseases and typical estimated values for \(\alpha\) for respiratory diseases range from 0.87 to 0.97. We used an \(\alpha\) value of 1 in these models.
In our model, individuals move from one node to another. A mobility matrix \(M\) where \(M(o,d)\) is the amount of individuals moving daily from origin \(o\) to destination \(d\). At each time step, and for each \((o,d)\) pair, we draw a force of infection for node \(i\) to account for mobility:
\[\begin{eqnarray} FOI_i &=& \left(1 - \sum_{j\neq i} p_{away} \frac{M_{i,j}}{H_i} \right) \cdot \beta_i(t) \frac{(I_1^{i} + I_2^{i} + I_3^{i})^\alpha}{H_i} + \\ && \sum_{j \neq i} \left(p_{away} \frac{M_{i,j}}{H_i} \cdot \beta_j(t) \frac{(I_1^j + I_2^j + I_3^j)^\alpha}{H_j} \right) \end{eqnarray}\]with \(p_{away}\) the percent of the time individuals that move spend away; \(p_{away} \approx 0.5\) in the case of commuting. \(H_i\) is the population of node \(i\). Then, the transition is:
\[\begin{equation} N_{S_i \longrightarrow I_1^{i}}(t) = Binom(S^i, 1 - \exp(-\Delta t \cdot FOI_i)) \end{equation}\]The model is implemented in python, just-in-time compiled to machine code using Numba.
We note several limitations to our work:
We truncated the initial run of these simulations on December 31, 2020. Our estimates therefore do not necesarily represent a total or final outbreak size in all scenarios. This truncation also has implications for the estimated number of hospitalizations and deaths, as many individuals have yet to reach their final outcome by December 31, 2020.
There remains considerable uncertainty around some of the key epidemiologic features of COVID-19, including the average duration of infectiousness and time to recovery or death. We have used commonly accepted and well supported estimates, in line with those recommended by the CDC, of these parameters and believe that they are appropriate for planning purposes.
We assume equal risk of infection and progression to hospitalization or death among all individuals within a county at a given time point. There is evidence of age-specific differences in clinical burden and perhaps in susceptibility to infection that are not considered here. Counties with older populations may experience a higher burden of clinical cases requiring hospitalization or ICU admission.
We assume \(R_e\), the effective reproductive number, to be constant in each scenario, other than changes due to onset of non-pharmaceutical interventions. As capacity for surveillance, contact tracing, and testing improve, and as the general public becomes increasingly aware of the outbreak and modifies their own behavior, we might expect that \(R_e\) will decrease dynamically, perhaps even as a function of the perceived COVID19 burden. As the outbreak continues, it will be possible to refine the scenarios considered here to better reflect the actual epidemic situation.
We do not explicitly model the role of asymptomatic infection when calculating the number of expected hospitalizations. All infectious individuals are considered at risk of hospitalization, though some may recover or die prior to hospitalization. A substantial asymptomatic burden may reduce the number of hospitalized cases.
Impact of broad scale non-pharmaceutical intervetions is based on the observed impact of such programs in 1918 as reported in (Bootsma and Ferguson 2007). We took parameter estimates on the effectiveness of interventions from Milwaukee in this study.
School closure impacts based on data from (Jackson et al. 2020), (Cauchemez et al. 2008) and (Litvinova et al. 2019)
Approximate generation time is based on data from (Bi et al. 2020)
Range of \(R_0\) selected to be consistent with the early stages of the Wuhan epidemic and is broadly consistent with estimates from Imperial College (https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-2019-nCoV-transmissibility.pdf), (Riou and Althaus 2020), (Majumder and Mandl 2020), (Zhao et al. 2020)
Assumption on ICU occupancy in the United States (Halpern and Pastores 2015)
Estimates of the transmission dampening parameter \(\alpha\) range from 0.87 to 0.97 for respiratory viruses from (Metcalf et al. 2011), (Word et al. 2012), (Finkenstadt, Bjornstad, and Grenfell 2002), (Becker and Grenfell 2017)
[ADD NAMES AND RECOMPILE PACKAGE AS NECESSARY]
This report would not be possible without the work of the COVID-19 Scenario Modeling Pipeline Working Group:
EPFL
Johns Hopkins Infectious Disease Dynamics
University of Utah
Developers Without Affiliation
Amazon Web Services
Becker, A.D., and B.T. Grenfell. 2017. “TSIR: An R Package for Time-Series Susceptible-Infected-Recovered Models of Epidemics.” PLOS One 12 (9).
Bi, Qifang, Yongsheng Wu, Shujiang Mei, Chenfei Ye, Xuan Zou, Zhen Zhang, Xiaojian Liu, et al. 2020. “Epidemiology and Transmission of COVID-19 in Shenzhen China: Analysis of 391 Cases and 1,286 of Their Close Contacts,” March. Cold Spring Harbor Laboratory. doi:10.1101/2020.03.03.20028423.
Bootsma, Martin C. J., and Neil M. Ferguson. 2007. “The Effect of Public Health Measures on the 1918 Influenza Pandemic in U.s. Cities.” Proceedings of the National Academy of Sciences 104 (18). Proceedings of the National Academy of Sciences: 7588–93. doi:10.1073/pnas.0611071104.
Cauchemez, Simon, Alain-Jacques Valleron, Pierre-Yves Boëlle, Antoine Flahault, and Neil M. Ferguson. 2008. “Estimating the Impact of School Closure on Influenza Transmission from Sentinel Data.” Nature 452 (7188). Springer Science; Business Media LLC: 750–54. doi:10.1038/nature06732.
Finkenstadt, B.F., O.N. Bjornstad, and B.T. Grenfell. 2002. “A Stochastic Model for Extinction and Recurrennce of Epidemics: Estimation and Inference for Measles Outbreaks.” Biostatistics 3 (4): 493–510.
Halpern, Neil A., and Stephen M. Pastores. 2015. “Critical Care Medicine Beds, Use, Occupancy and Costs in the United States: A Methodological Review.” Critical Care Medicine 43 (11). Critical Care Medicine: 2452–9. doi:10.1097/CCM.0000000000001227.
Jackson, Michael L, Gregory R Hart, Denise J McCulloch, Amanda Adler, Elisabeth Brandstetter, Kairsten Fay, Peter Han, et al. 2020. “Effects of Weather-Related Social Distancing on City-Scale Transmission of Respiratory Viruses,” March. Cold Spring Harbor Laboratory. doi:10.1101/2020.03.02.20027599.
Litvinova, Maria, Quan-Hui Liu, Evgeny S. Kulikov, and Marco Ajelli. 2019. “Reactive School Closure Weakens the Network of Social Interactions and Reduces the Spread of Influenza.” Proceedings of the National Academy of Sciences 116 (27). Proceedings of the National Academy of Sciences: 13174–81. doi:10.1073/pnas.1821298116.
Majumder, Maimuna, and Kenneth D. Mandl. 2020. “Early Transmissibility Assessment of a Novel Coronavirus in Wuhan, China.” SSRN Electronic Journal. Elsevier BV. doi:10.2139/ssrn.3524675.
Metcalf, C.J.E., C. V. Munayco, C. Viboud, G. Chowell, B.T. Grenfell, and O.N. Bjornstad. 2011. “Rubella Metapopulation Dynamics and Importance of Spatial Coupling to the Risk of Congenital Rubella Syndrome in Peru.” Journal of the Royal Society Interface 8 (56): 369–76.
Riou, Julien, and Christian L. Althaus. 2020. “Pattern of Early Human-to-Human Transmission of Wuhan 2019-nCoV,” January. Cold Spring Harbor Laboratory. doi:10.1101/2020.01.23.917351.
Word, D.P., D.A. Cummings, D.S. Burke, S. Iamsirithaworn, and C.D. Laird. 2012. “A Nonlinear Programming Approach for Estimation of Transmission Parameters in Childhood Infectious Disease Using a Continuous Time Model.” Journal of the Royal Society Interface 9 (73): 1983–97.
Zhao, Shi, Qianyin Lin, Jinjun Ran, Salihu S. Musa, Guangpu Yang, Weiming Wang, Yijun Lou, et al. 2020. “Preliminary Estimation of the Basic Reproduction Number of Novel Coronavirus (2019-nCoV) in China, from 2019 to 2020: A Data-Driven Analysis in the Early Phase of the Outbreak.” International Journal of Infectious Diseases 92 (March). Elsevier BV: 214–17. doi:10.1016/j.ijid.2020.01.050.